The paper, ‘Towards a Foundation Purchasing Model: Pretrained Generative Autoregression on Transaction Sequences’, presents a generative pretraining method to obtain contextualised embeddings of financial transactions. The model, which outperforms state-of-the-art self-supervised methods, is based on machine learning and is used in modern financial systems for fraud detection. The researchers pretrained an embedding model using data from 180 issuing banks containing 5.1 billion transactions. The model significantly improves value detection rate at high precision thresholds and transfers well to out-of-domain distributions.
Publication date: 4 Jan 2024
Project Page: https://doi.org/10.1145/3604237.3626850
Paper: https://arxiv.org/pdf/2401.01641
